Richard Williams

The emergence of AIs has allowed a new type of epistocratic governance to emerge: AIocracy. The democracy/epistocracy debate largely considers the plausibility of weighted votes in elections. In contrast, AIocracy empowers artificial intelligence (AI) to weigh voices during online deliberations. This paper uses a Condorcetian model of democracy to show how the theoretical benefits of AIocracy translate into actual epistemic harms. In particular, AI domination—as a technology-driven type of domination—allows an alien type of algorithmic rationality to compromize truth-oriented politics with an engagement-oriented deliberative environment that manipulates particular people and the general deliberative process itself. Most significantly, AIs amplify salient but simplistic narratives that intentionally maximize engagement but unintentionally maximize emotions, which unintentionally minimize the independence, competence, and sincerity of voters. So, AI domination threatens to invert the tendency of deliberation to cultivate our Condorcetian characteristics and compromize the competence of democracy itself.

The emergence of AIs has allowed a new type of epistocratic governance to emerge: AIocracy. The democracy/epistocracy debate largely considers the plausibility of weighted votes in elections. In contrast, AIocracy empowers artificial intelligence (AI) to weigh voices during online deliberations. This paper uses a Condorcetian model of democracy to show how the theoretical benefits of AIocracy translate into actual epistemic harms. In particular, AI domination—as a technology-driven type of domination—allows an alien type of algorithmic rationality to compromize truth-oriented politics with an engagement-oriented deliberative environment that manipulates particular people and the general deliberative process itself. Most significantly, AIs amplify salient but simplistic narratives that intentionally maximize engagement but unintentionally maximize emotions, which unintentionally minimize the independence, competence, and sincerity of voters. So, AI domination threatens to invert the tendency of deliberation to cultivate our Condorcetian characteristics and compromize the competence of democracy itself.

The use of artificial intelligence (AI) to transmit information in democratic societies presents a new problem for scientific objectivity. This paper extends objectivity from knowledge production to knowledge transmission. It analyses how an overlooked type of agent—knowledge intermediaries—produces an overlooked type of trust—external trust. In practice, citizens rely on intermediaries to transmit trustworthy information. This paper argues that centralised techno-legal approaches, privileging AIs as epistemically superior intermediaries, risk epistemic harm to human intermediaries. In contrast, a decentralised intermediaries approach that values contestation among human intermediaries is a better way to transmit trustworthy information and resist misinformation.

Epistemic democrats indirectly evaluate democratic decisions by directly evaluating the inputs into the election. However, the fundamental problem of measurement in the philosophy of science shows that procedures are often as difficult to evaluate as outcomes. This paper brings this highly refined framework into political philosophy to show that epistemic democrats face an analogous ‘fundamental problem of evaluation’. This cross-fertilization of political philosophy with the philosophy of science shows that the quality of democratic mechanisms and their inputs regarding their ability to track the truths of justice is as difficult to evaluate as the quality of the resulting decisions themselves.

Epistemic democrats typically argue that widespread public competence can empower democratic states to produce the correct decisions more effectively than antidemocratic alternatives. In reaction, this paper shows that epistemic democrats are too insensitive to a fundamental fact of representative democracies: the democratic choice of policy is mediated through a democratic choice of politician. Epistemic democrats neglect that party politicians potentially spoil the epistemic benefits of widespread public competence. Firstly, politicians must compete with each other for votes during elections. Secondly, politicians should compromise with each other to protect those they represent from the bad apples in the legislature. Politicians, as elected representatives with democratic integrity, have a profession-specific obligation to resist the bad apples, even if they must sacrifice their personal integrity in the process. They must compromise on promoting the correct decisions to gain critical political alliances and electoral support. Once political theorizing recognizes the significance of party politicians and their obligations more fully, public deliberation can be modelled as a compromise-discovery process: public deliberation can enable politicians to know which moral compromises will gain the alliances and votes necessary to resist the bad apples.